GitHub Copilot: Dynamic Workflows in Copilot CLI and the Copilot App
GitHub CopilotGitHub introduced dynamic workflows in public preview for Copilot CLI, the GitHub Copilot app, and the GitHub Copilot SDK. A dynamic workflow is a program that defines a task in code, combining automated steps with one or more agents that run sequentially, in parallel, or both. This gives complex multi-agent work repeatable stages, checkpoints, and observability, and it is available on all Copilot plans.
Key Takeaways
- Dynamic workflows put orchestration in code, so the same stages run every time while agents handle only the steps that need judgment.
- Steps can run sequentially, in parallel, or both, and structured results pass between stages for dependable multi-agent pipelines.
- Workflows can include subagent verification and user checkpoints, pausing for review before resuming a long run.
- Strong fits include parallel file review, release checks, codebase sweeps, and incident investigation, while simple tasks stay in chat mode.
- The feature spans Copilot CLI, the Copilot app, and the Copilot SDK, and runs inside a Copilot extension with access to its extensibility APIs.
- It is a public preview on all Copilot plans, with no setup in the app and
--experimentalor/experimental onneeded in the CLI.
Orchestration Defined in Code
GitHub introduced dynamic workflows in public preview across Copilot CLI, the GitHub Copilot app, and the GitHub Copilot SDK. The idea is to let developers define an orchestration in code so that complex, multi-agent work gains the reliability and observability it needs. A dynamic workflow is a program that describes how a task is carried out. It mixes automated steps with the work of one or more agents, and those steps can run one after another, in parallel, or in a combination of both.
The key distinction is who decides the structure. In a dynamic workflow, the steps, the moments agents get involved, and the way their results are used are all written in code. Agents handle the parts that call for analysis or judgment, while the surrounding process stays deterministic. The program lives inside a GitHub Copilot extension, so it can use Copilot's extensibility APIs.
What a Workflow Can Do
GitHub gives the example of investigating a service incident. A workflow can collect logs and telemetry, assign independent agents to analyze different systems, and combine their structured findings into a timeline and root-cause report, with the same steps running every time. More broadly, dynamic workflows can run commands, use tools, or call other services, divide goals into parallel tasks, pass structured results between stages, have subagents verify findings, request user input, and pause at checkpoints for review before resuming.
When It Makes Sense
GitHub recommends dynamic workflows for reusable processes, or for single tasks that need clear stages, checks, or limits. Suggested use cases include running release checks with agent assessment while pausing for review, reviewing many changed files in parallel, using code to find unresolved comments with dual-model verification, sweeping a codebase for patterns, researching changes before implementation, and kicking off long runs that may need to be paused.
For quick answers or simple changes, standard chat mode is still the better fit.
Availability and Setup
Dynamic workflows are available on all Copilot plans. In the GitHub Copilot app they require no setup. In Copilot CLI, developers enable experimental features with the --experimental option, or with /experimental on in an interactive session. Developers can ask Copilot to create workflows for processes they run often, or ask which workflows already exist. Feedback goes through /feedback in Copilot CLI. The feature is in public preview and subject to change.